Journal of Rock Mechanics and Geotechnical Engineering 14 (2022) 210e218

Contents lists available at ScienceDirect

Journal of Rock Mechanics and

Geotechnical Engineering
journal homepage: www.jrmge.cn

Full Length Article

Estimation of thermal conductivity of cemented sands using thermal

network models
Wenbin Fei, Guillermo A. Narsilio*
Department of Infrastructure Engineering, The University of Melbourne, Parkville, Australia

a r t i c l e i n f o a b s t r a c t

Article history: Effective thermal conductivity of soils can be enhanced to achieve higher efficiencies in the operation of
Received 29 March 2021 shallow geothermal systems. Soil cementation is a ground improvement technique that can increase the
Received in revised form interparticle contact area, leading to a high effective thermal conductivity. However, cementation may
17 June 2021
occur at different locations in the soil matrix, i.e. interparticle contacts, evenly or unevenly around
Accepted 20 August 2021
Available online 1 November 2021
particles, in the pore space or a combination of these. The topology of cementation at the particle scale
and its influence on soil response have not been studied in detail to date. Additionally, soils are made of
particles with different shapes, but the impact of particle shape on the cementation and the resulting
Keywords:
Network
change of effective thermal conductivity require further research. In this work, three kinds of sands with
Cementation different particle shapes were selected and cementation was formed either evenly around the particles,
Computed tomography or along the direction parallel or perpendicular to that of heat transfer. The effective thermal conductivity
Ground improvement of each sample was computed using a thermal conductance network model. Results show that dry sand
Sands with more irregular particle shape and cemented along the heat transfer direction will lead to a more
efficient thermal enhancement of the soil, i.e. a comparatively higher soil effective thermal conductivity.
Ó 2022 Institute of Rock and Soil Mechanics, Chinese Academy of Sciences. Production and hosting by
Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/
licenses/by-nc-nd/4.0/).

1. Introduction influence of cementation in heat transfer, and to reduce the number

Effective thermal conductivity (leff) is one of the key soil pa-
rameters required for shallow geothermal engineering. Moisture
content is the main factor influencing the soil thermal conductivity
since water thermal conductivity (0.596 W/(m K)) is more than 20
times that of air (0.025 W/(m K)) (Young et al., 1996). The thermal
conductivity of the solid particles is one order of magnitude higher
than that of water. Therefore, soil porosity or bulk density, indi-
cating the fraction of solid particles, also has an influence on leff.
Unless the soils are completely saturated (high moisture content),
or work in arid environments (low moisture content), their mois-
ture content typically varies and is hard to be controlled; and
porosity or bulk density can only be typically enhanced in engi-
neered fills or (typically) in up to very shallow depths (<1 m).
Therefore, innovative approaches are required to enhance the
thermal conductivity of soils. Deep mixing and grouting (soil
cementation) may constitute one such approach. To understand the
* Corresponding author.
E-mail addresses: wenbinfei@outlook.com (W. Fei), narsilio@unimelb.edu.au
(G.A. Narsilio).
Peer review under responsibility of Institute of Rock and Soil Mechanics, Chi-
nese Academy of Sciences.
of influencing factors (e.g. moisture and compaction), here we
study dry geomaterials as a first step since soils in arid environ-
ments could be considered mostly dry from their heat transfer
mechanisms point of view.
Heat transfer in dry soils at a temperature lower than 700
C (El
Shamy et al., 2013; Asakuma et al., 2016) is mainly due to con-
duction through three paths: particles themselves, interparticle
contacts and near-contacts (Yun and Santamarina, 2008; Fei et al.,
2019a). Grouting, a traditional ground improvement method, can
cement pre-existing particles together to achieve a superior
bearing capacity for foundations. It also increases the interparticle
contact area and thus offers the extra benefit of superior heat
transfer for the ground since interparticle contacts are a primary
heat transfer path (Yun and Santamarina, 2008; Narsilio et al.,
2010a, b). However, the enhancement of soil thermal conductivity
for a more efficient geothermal system has not been paid enough
attention to. Cementation can be achieved by injecting cement
grout into the ground (deep mixing and grouting) or by microbially
induced calcite precipitation (MICP) technique as an alternative
(Wang et al., 2020). The two different approaches may lead to
cementation at different locations within the soil matrix or to
cementation anisotropy (preferential cementation growth di-
rections), resulting in different leff values. However, related studies

https://doi.org/10.1016/j.jrmge.2021.08.008
1674-7755 Ó 2022 Institute of Rock and Soil Mechanics, Chinese Academy of Sciences. Production and hosting by Elsevier B.V. This is an open access article under the CC BY-
NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
 W. Fei, G.A. Narsilio / Journal of Rock Mechanics and Geotechnical Engineering 14 (2022) 210e218
have not been conducted. Additionally, since particles in natural
sands have distinct shapes, the investigation of how particle shape
affects the enhancement of leff due to cementation is also
necessary.
One of the essential works of studying heat transfer in the soils
is to calculate or measure the leff accurately and efficiently.
Although laboratory results are reliable, testing requires relatively
large non-destructive samples that are sometimes difficult or costly
to obtain, as for sandy soils in particular. It is more challenging to
build an empirical equation for predicting the thermal conductivity
of these sandy soils since more samples are required. Another
limitation in the empirical derivation is that the microstructure
information controlling the macroscopic (or engineering) heat
transfer is limited to parameters such as mean particle size, void
ratio, and mean coordination number. However, the inaccessibility
of local microstructural information such as interparticle contact
area and its distribution constrains the fundamental understanding
of the mechanisms of heat transfer in natural soils (cemented or
uncemented). Even with access to local microstructural informa-
tion, this information and leff from numerous samples are required
to build a reliable artificial intelligence (AI) predictive model
(Zhang et al., 2020; Fei et al., 2021a).
The discrete element method (DEM) has been used to simulate
heat transfer in spherical granular media with the convenience of
extracting the particle size and coordination number of each par-
ticle (Chaudhuri et al., 2006; Moscardini et al., 2018). Nevertheless,
it may not be applicable for natural soils in which soil particles
show complex shapes even though spheres can be clumped
together to imitate an irregular particle. To address this limitation,
an X-ray computed tomography (XCT) based level-set DEM (de
Macedo et al., 2018) has been developed, but the current version
has no functionality of simulating heat transfer. Alternatively, XCT
based finite element simulation discretises irregular particles into
meshes, in which energy balance equations are solved (Narsilio
et al., 2010b; Fei et al., 2019b). Even though the finite element
method (FEM) simulates local heat transfer inside the particle and
at interparticle contacts, without further XCT image post-
processing during the simulation, FEM cannot deal with particle
volume effect and may lead to an overestimation of leff (Persson
et al., 2005; Wiebicke et al., 2017).
Recently, network models have been introduced to simulate
transport properties directly using XCT images but widely used for
fluid flow rather than heat transfer (Gostick et al., 2016; van der
Linden et al., 2016). A network model presents a sample as a web
of nodes and edges. In a pore network, a node presents a void while
an edge indicates a pore throat that connects two neighbouring
pores. Our research group has pioneered the development of an
XCT image based thermal conductance network model (TCNM) that
Fig. 1. The computed tomography (CT) images of three sands: (a) Glass beads, (b) Ottawa sand, and (c) Angular sand.
211
treats particles as nodes linked by edges presenting interparticle
contacts or near-contacts (Fei et al., 2019a; van der Linden et al.,
2021). TCNM first introduces a greyscale intensity penalty factor
for mitigating the overestimation of contact area and leff due to the
partial volume effect. It also shows high computational efficiency
by detecting the adjacency of solid and void phases after watershed
segmentation, which is hinted at by the dual graph of Voronoi di-
agram and Delaunay triangulation (Aurenhammer et al., 2013). The
TCNM approach and associated tools such as three-dimensional
(3D) particle shape descriptor and complex network theory (i.e.
graph theory) can achieve richer valuable microstructural param-
eters (Fei and Narsilio, 2020). Hence, this combined approach re-
solves the hypothesis that thermal front propagation does not
merely depend on global porosity (Yun and Evans, 2010). The
available new network features are important microstructural pa-
rameters that can boost more reliable heat transfer models in either
a theoretical or AI approach.
This paper aims to investigate the extra benefit of cementation to
enhance not only the geomechanical response of the ground, but
also the thermal conductivity of soils. Three natural sands were
selected for scanning and cementation was simulated by dilating the
solid phase in XCT images along different directions. The leff of the
cemented sands was computed using TCNM. With these results, the
impact of cementation on leff for the three sands were compared
followed by the impact of cementation anisotropy on leff .
2. Materials
2.1. Uncemented sands
Glass beads, Ottawa sand and angular sand were selected in this
work since the complexity of their particle shape increases from
spherical, near-spherical to irregular, as shown in Fig. 1. Glass beads
and Ottawa sand are two common research materials used in
geotechnical experiments. Glass beads are made of silica, while
both Ottawa sand and angular sand mainly consist of quartz. Their
equivalent particle sizes (Table 1) derived from particle volume are
similar: 0.6 mm for glass beads, 0.76 mm for Ottawa sand, and
0.68 mm for angular sand. XCT images can be freely accessed by
readers in Fei et al. (2021b).
2.2. Cemented sands
Grouting is an artificial surface nucleating procedure to intro-
duce new solids in the pore space of soils. Natural nucleation is also
widely observed in deep geothermal projects (power generation)
when steam and fluid are circulated in an open aquifer where
mineral precipitation induced by variations of temperature and
212 W. Fei, G.A. Narsilio / Journal of Rock Mechanics and Geotechnical Engineering 14 (2022) 210e218

Table 1 rectangles from Fig. 2a and b is presented in Fig. 3. While solid voxels
Particle size (mm) of the studied granular materials. in Fig. 3a are in light grey and real void voxels are in black, some void
Sample Radius from sieve analysis Equivalent ball radius derived voxels at the interparticle contacts are partially filled with grey
from particle volume after CT during XCT scanning. These partially filled void voxels were incor-
reconstruction rectly grouped as solid in the binary image after threshold seg-
Average radius Radius range Average radius Radius range mentation, as shown in Fig. 3b, which is called partial volume effect
Glass beads 0.3 0.25e0.35 0.3 0.2e0.4 (Wiebicke et al., 2017). In the present work, this overestimated solid
Ottawa sand 0.36 0.3e0.42 0.38 0.29e0.47 phase at the interparticle contact was considered as local cemen-
Angular sand 0.44 0.3e0.59 0.34 0.19e0.5 tation (sample D0) to imitate the cementation bridge between soil
particles due to mineral precipitation (Wang et al., 2020). Hence, 11
pressure occurs. To imitate the surface nucleation that generates cemented samples for each kind of sand were generated to simulate
new solids on the surface of pre-existing particles, a cubic sub- surface cementation in this paper (D0eD10).
sample (350 pixels in length, and pixel size is 13 mm) from the XCT MICP is an alternative way to achieve cemented sands (Wang
images of each sand was first cropped, as shown in Fig. 2a. It is et al., 2020). The underground fluid flow could control the chemi-
noticeable that XCT images are two-dimensional (2D) images in cal alteration (cementation/mineralisation or dissolution) in
sequence with an interval of a pixel size and stacking them can different approaches (Lichtner, 1988). For example, the propagation
reconstruct a digital 3D sample. Next, Otsu threshold segmentation of reaction fronts can lead to a banded or layered pattern such as
(Otsu, 1979) was utilised to identify the solid and void phases, marble. The variations of temperature and pressure along the
resulting in a binary image with solid in black and void in white, as streamlines could result in gradient reactions. These cases can
shown in Fig. 2b. After that, 3D dilation was applied to the pre- result in anisotropic cementation along different directions as
existing solid phase in the binary image to generate the extra shown in the yellow windows in Fig. 4. The two particles in window
solids around the pre-existing particles, as shown in Fig. 2c. To A are connected due to horizontal mineral precipitation (Fig. 4a),
better visualise the nucleation, the new solids were rendered as while they are still separated because of vertical mineral precipi-
white around the original greyscale particles in Fig. 2d. This ‘virtual’ tation (Fig. 4b). In contrast, the two particles in window B keep the
dilation was varied from 1 pixel to 10 pixels in this work, and the small gap after horizontal mineral precipitation (Fig. 4a), while they
obtained 10 cemented samples of varying cementation degree for become in contact resulting from vertical mineral precipitation
each sand were labelled as D1eD10, respectively. (Fig. 4b). The directional cementation can alert the preferential heat
Noticeably, solids could be over-generated during threshold transfer path and intensify the thermal anisotropy of the sands. The
segmentation. The amplification of the local images in the yellow procedures of producing the samples cemented in all directions

Fig. 2. Approach to achieve surface cemented Ottawa sand. The original greyscale CT image (a) was threshold segmented into a binary image (b) followed by a 2-pixel dilation of the
solid phase (c), which results in the extra white pixels in (d) presenting grout.

Fig. 3. Overestimation of the contact area due to partial volume effect: (a) Original greyscale image; and (b) Binary image after threshold segmentation.
 W. Fei, G.A. Narsilio / Journal of Rock Mechanics and Geotechnical Engineering 14 (2022) 210e218 213

Fig. 4. Mineral precipitation in (a) horizontal and (b) vertical directions. A 2-pixel dilation was applied.

have been illustrated in Fig. 2. A similar approach but in horizontal
and vertical directions was applied to the original sands to generate
directionally cemented samples. One-pixel directional cementation
was iterated 10 times for each kind of sand. Hence, 10 horizontally
cemented samples and 10 vertically cemented samples were
generated for each kind of sand.
3. Thermal conductance network model
3.1. Network construction
The overestimated contact area shown in Fig. 3b leads to the
overestimation of leff since interparticle contact is a primary heat
transfer path, especially in dry sands (Yun and Santamarina, 2008).
The overestimation of leff cannot be mitigated using traditionally
simulation methods such as the FEM. To address this barrier, a
TCNM was developed by our research team (Fei et al., 2019a; van
der Linden et al., 2021) and implemented in the simulation of
heat transfer in this paper. TCNM is an XCT and network based
simulation tool, thus it provides the benefit of post-processing
images during the computation.
An XCT image stack (Fig. 5a) was used to reconstruct 3D digital
sand shown in Fig. 5b after Otsu threshold segmentation. A network is
a web of nodes and edges, and in a thermal network, nodes were used
to present particles while edges were used for interparticle contact or
near-contacts to cover the main heat transfer paths in dry granular
materials (a contact network typically only considers real contacts as
edges, not near-contacts). Consequently, watershed segmentation
was employed to split the integrated solid phase in Fig. 5b into indi-
vidual particles with unique identifiers (IDs) and rendered with
random colours in Fig. 5c. For each particle, a black node was assigned
to its coordinate centre, as shown in Fig. 5d. To detect interparticle

Fig. 5. Procedures of simulating heat transfer in the thermal conductance network model (TCNM).
214 W. Fei, G.A. Narsilio / Journal of Rock Mechanics and Geotechnical Engineering 14 (2022) 210e218

contact, boundary voxels of each particle were recognised. Next, if the X g  2 X 1

boundary voxels of a particle i were shared with particle j, they
contact
Cgap ¼ cl ¼ lv Lv g (4)
l
L
l l
belonged to interparticle contacts and were presented as red edges, as
shown in Fig. 5e. To identify near-contacts efficiently, watershed
where lv is the thermal conductivity of the void phase, Lv is the
segmentation was also applied to the void phase as shown in grey- g
pixel size of XCT images, and Ll is the length of each blue near-
scale in Fig. 5d. After that, if particle boundary voxels bordered on a
contact cylinder in Fig. 5e and computed using the distance be-
pore, particle ID and pore ID were grouped as a particle pore
tween boundary voxels in neighbouring particles. The maximum
connection illustrated as a purple arrow in Fig. 5d. Then, potential
value of Lgl is ε, which is also the cut-off to determine the effective
near-contacts shown as blue edge could be efficiently identified. The
zone for particleefluideparticle heat transfer.
near-contacts are essentially a group of gaps between boundary
voxels of two neighbouring particles shown as blue cylinders in
(3) Thermal conductance at interparticle contact when two
Fig. 5e. If the length of the shortest blue cylinders was smaller than a
particles share the same boundary voxels (purple and green
cut-off range ε, a near-contact was built in the thermal network as
cylinders between particles j and k in Fig. 5e):
blue edges. In addition, all blue cylinders with lengths smaller than ε
will be later used to compute the thermal conductance of the near-
contact. P
kAC k n An
v
By now, the thermal network has only presented the topology contact
Cinterparticle ¼ ls þ C gap ¼ ls v
þ C gap (5)
inside sand. To simulate heat transfer using the thermal network,
LC 3L
particles and contacts were treated as conductors connecting the where AC is the interparticle contact area computed as the sum
top and bottom boundaries of the whole assembly, as shown in of the area Avn of the shared boundary voxels. To overcome the
Fig. 5f. Therefore, Fourier’s law can be used to calculate the required aforementioned partial volume effect, Avn was corrected using Eq.
heat flux transferring through the sample: (6) by applying penalty parameter s to each local greyscale gn of
X X   the shared voxels and their maximum greyscale gmax contact . k is
Qij ¼ Cij Ti  Tj (1) another penalty coefficient to reduce the overestimation because
i/j i/j
of neglecting roughness at interparticle contacts. It was set as
0.75 for the original greyscale XCT images since the work of
where Qij is the heat flux transferring from node i to node j, which is
Askari et al. (2015) showed that a 25% overestimation could
induced by a temperature difference between the ends of a
occur if the images were left untreated after threshold seg-
conductor with conductance Cij . One end of the conductor is node i
mentation. LC is the length of the interparticle contact cylinder,
with temperature Ti while the other end is node j with temperature
which is assumed to be 3Lv (van der Linden et al., 2021) refer-
Tj . Therefore, one of the keys to compute leff using TCNM is to
ring to the studies of Bauer and Schlunder (1978) and Shapiro
calculate the Cij between nodes i and j.
et al. (2004):
 s
3.2. Computation of thermal conductance gn
Avn ¼ contact
L2v (6)
gmax
Green, purple and blue cylinders in Fig. 5e were used to calcu-
late the thermal conductance in particles, interparticle contacts and X g  2 X 1
near-contacts. The thermal conductance C of a cylinder could be C gap ¼ Cl ¼ ln L v g (7)
simply computed using the following equation: l
L
l l

Once the local thermal conductance was available, the effective

C ¼ lA=L (2)
thermal conductance between two particles was calculated as fol-
where l is the thermal conductivity, A is the cross-section area of lows for the contact condition that a gap smaller than ε exists be-
the cylinder, and L is the length of the cylinder. Then, the main task tween two neighbouring particles:
of computing the thermal conductance for different cylinders was 0 11
to achieve the associated l; A and L. The following expressions were 1 1 1
used to compute local thermal conductance: Cij ¼ @ P þ contact þ P A (8)
Ci Cijgap Cj

(1) Particle conductance (green cylinders in Fig. 5e): For the interparticle condition, the effective thermal conduc-
tance is computed as follows:
.
!1
A p cV P LP 1 1 1
C p ¼ ls p ¼ ls (3) Cjk ¼ þ contact þ P (9)
L LP CjP Cjk C
interparticle k

where ls is the thermal conductivity of the solid, LP is the distance

between particle centroid and the corresponding interparticle
contact or near-contact, and Ap is the cross-section area of particle
cylinder and computed using a fraction coefficient c to consider
part of the particle volume V P to avoid the overestimation of the
heat transfer inside the particle. For sphere packings, 0.25 was
assigned to c (Yun and Evans, 2010).
(2) Thermal conductance between two particles with a gap (blue
cylinders between particles i and j in Fig. 5e), which is the
g
sum of thermal conductance Cl of a group of gap cylinders:
3.3. Computation of effective thermal conductivity
According to the equations in the previous section, the input
parameters in TCNM are summarised in Table 2. For mineral grains
in soil, their thermal conductivity is usually set as 3 W/(m K) while
0.025 W/(m K) for air (Young et al., 1996). The determination of c
and ε is based on: (1) the calibration of the leff of same sphere
packings from our TCNM with that from sphere packing thermal
network model developed by Yun and Evans (2010); and (2) the
 W. Fei, G.A. Narsilio / Journal of Rock Mechanics and Geotechnical Engineering 14 (2022) 210e218 215

best match with theoretical and experimental results shown in
Section 5.1. A quarter of particle volume was used to compute the
leff of sphere packings, i.e. c was 0.25 in Eq. (3) according to Yun
and Evans (2010), and in this work, a value of 0.23 was assigned
to c to best match the theoretical and experimental leff of natural
sands made of irregular particles. The value of the cut-off range ε of
near-contact in a sphere packing was recommended as the har-
1
monic mean, i.e. 0:5ðri1 þ rj1 Þ of two neighbouring particle radii
ri and rj (Yun and Evans, 2010), we used the average, i.e. 0.5ðri þrj Þ;
for natural sand after calibration processes (van der Linden et al.,
2021). The original greyscale XCT images were used to compute
the leff of original sands, and k ¼ 0.75 and s ¼ 10 were imported
to TCNM to reduce the overestimation during the transmission
from greyscale to binary images (Fig. 5a and b) (van der Linden
et al., 2021). As binary images were used to compute the leff of
cemented samples, penalty was not required for k and s to deal with
the binary images. Hence, they were set as 1 and 0 for the
computation of leff for cemented samples, respectively.
After setting the required input parameters, boundary temper-
ature Ttop ¼ 293 K and Tbot ¼ 292 K were imposed to the top and
bottom nodes in the thermal network, respectively, as shown in
Fig. 5f, to generate a thermal gradient. Next, a Python library
OpenPNM (Gostick et al., 2016) was used to compute the temper-
ature and heat flux at each node by importing the local conduc-
tance and boundary temperatures to Eq. (1) followed by successive
iteration. Finally, the effective thermal conductivity of the sample
can be computed as
1
P models, which are larger than the TCNM and experimental results
Qij
leff ¼  A 
Ttop  Tbot L Shtrikaman (HS-L) model. leff obtained from TCNM shows a good
P agreement with experimental results in Fig. 6b. In contrast, FEM
where Qij is the sum of the heat flux of all the nodes on the cross-
section. In this paper, both the top and bottom planes were selected
Table 2
Input parameters in the thermal conductance network model.
Parameters Description Value
ls Solid thermal conductivity 3 W/(m K)
lv Void thermal conductivity 0.025 W/(m K)
X Fraction of particle volume 0.23 after comparison, as shown in
Section 5.1
ε Cut-off distance of the near- 0.5r/pixel size, where r is the average
contact conductance radius, r ¼ 0:5ðri þ rj Þ
K Surface roughness penalty 0.75 for greyscale images, and 1 for
binary images
T Particle volume effect 10 for greyscale images, and 0 for
penalty binary images
to compute their own leff and their average was treated as the leff
of the sample.
4. Results and discussion
4.1. Model verification
For the verification of the TCNM, the leff values of glass beads
and angular sand at rest condition as well as Ottawa sand under
four stages of uniaxial compression (0 MPa, 2 MPa, 6.1 MPa, and
10.2 MPa) were computed using both TCNM and FEM implemented
in COMSOL Multiphysics (COMSOL AB, 2020). The steps of calcu-
lating leff of natural sands based on XCT images using the FEM is
detailed elsewhere (Fei et al., 2019b). The leff values of Ottawa sand
and angular sand at rest were measured by the authors using a
thermal needle probe (KD2 Pro thermal properties analyser from
Decagon Devices, Inc.) following the standard ASTM D5334e14
(2014). The leff results measured by Yun and Santamarina (2008),
Narsilio et al. (2010b), and Tarnawski et al. (2011) and calculated
using the theoretical formulations are summarised in Table 3, and
comparison was also conducted, as shown in Fig. 6.
To show the difference between the theoretical formulations,
the variation of leff with porosity from 0 to 1 is shown in a semi-log
plot in Fig. 6a. leff is the same as the solid thermal conductivity
when the porosity is 0, and same as the void thermal conductivity
when the porosity is 1. Results from FEM, TCNM and experiments
sit within the theoretical results. Specifically, FEM results are
located between the self-consistent (SC) and geometric mean (GM)
located between GM and the lower bound of the Hashin-
(10)
results are larger than experimental results, presenting obvious
overestimations, especially when the porosity is small. The
decrease of porosity means more contacts and stronger possibility
of overestimating interparticle contact area, leading to the over-
estimation of leff . Similarly, in the uniaxial compression tests of
Ottawa sands, the overestimation increases when the sample is
subjected to a larger loading.
4.2. Effective thermal conductivity of surface cemented sands
Apart from applying an external loading to enhance the leff of
original sands, as shown in Fig. 6, the interparticle contact area can
be directly increased by cementation to obtain a larger leff . For each
sand, the original uncemented sample and 11 cemented samples
were prepared and their leff values were computed using TCNM.
For Ottawa sand, its leff increases from 0.29 W/(m K) to 0.44 W/(m

Table 3
Theoretical models for predicting effective thermal conductivity.

Model Equation Researchers

Series !1 DeVera and Strieder (1977)

P
N f
leff ¼ i
i ¼ 1 li
Parallel P
N
leff ¼ fi li
i¼1
Geometric mean (GM) leff ¼ lfii Sass et al. (1971)
 
Hashin-Shtrikaman (HSeU: upper bound; 3f2 ðl2  l1 Þ Hashin and Shtrikman (1962)
leff ¼ l1 1 þ (Upper: 1 ¼ solid, 2 ¼ pore;
HS-L: lower bound) 3l1 þ f1 ðl2  l1 Þ
Lower: 1 ¼ pore, 2 ¼ solid)
Self-consistent (SC) !1 Hill (1965), Tarnawski et al. (2002)
1 1n n
leff ¼ þ
3 2leff þ ls 2leff þ ls

Note: li is the thermal conductivity of the phase with volume fraction fi , and n is the porosity.
216 W. Fei, G.A. Narsilio / Journal of Rock Mechanics and Geotechnical Engineering 14 (2022) 210e218
Fig. 6. Comparison of effective thermal conductivity among different resources from FEM, TCNM, experiments and theoretical models: (a) The theoretical results in log scale with
porosity ranging from 0 to 1; and (b) A subplot of (a) with a smaller range of porosity and leff in normal scale. Note that the colours of markers present different sands while the
shapes of markers indicate how the leff was obtained.
Fig. 7. Comparison of the effects of cementation and compression on leff from the
thermal conductance network model.
K) after applying 10.2 MPa axial stress, as shown in Fig. 7 (particle
or cementation breakages are not accounted for). In contrast, extra
solid caused by cementation only occupies the gaps between real
interparticle contacts and makes the point-to-point contacts
become surface-to-surface contacts in D0 sample (i.e. partially fil-
led void voxels become solid phase in Fig. 3). The increased contact
area enhances the leff of the original Ottawa sand from 0.288 W/(m
K) to 0.588 W/(m K). In D2 sample, a thin surface cementation was
generated, as shown in Fig. 2d, which increases the leff 150%. One
can achieve a reduction in porosity (compaction or densification) in
a site by the traditional preloading ground improvement technique,
but since 10 MPa stress is not easy to be achieved using traditional
preloading on an engineering site, it follows that cementation could
be a more accessible approach to enhance the leff of soils in practice
(and would not prompt particle breakage).
In the original sands, angular sand has the largest porosity and
the least solid among the three sands, as shown in Fig. 8. However,
the porosity decreases and extra solid volume increases at the
fastest speed in angular sand, which later leads to a larger leff than
that of glass beads and Ottawa sand at the same porosity.
The cemented samples D0eD10 for each sand are related to the
cementation level generated by pixel dilation. The correlation be-
tween cementation level and the ratio of the leff of each cemented
sand to that of the original sand is presented in Fig. 9a. Angular sand
has the highest increase in leff at each cementation level, which
indicates that the leff of a sand consisting of a more complex par-
ticle shape may be more prone to be thermally enhanced. The faster
increment of leff in angular sand than that in others is owing to the
quicker increase of its average coordination number (CNave), as
shown in Fig. 9b. The coordination number corresponds to the
interparticle contact rendered as red edges in Fig. 5e, acting as an
essential heat transfer path in granular materials (Yun and
Santamarina, 2008).
4.3. Effective thermal conductivity of directional cementation
Instead of injecting cement into the soil to reinforce the ground,
MICP is another engineering method to enhance cementation be-
tween sand particles (Wang et al., 2020), and the cementation may
be anisotropic due to the fluid flow direction and pressure or
temperature gradient. The anisotropic cementation can intensify
the thermal anisotropy of the sands since the internal preferential
heat transfer path could be changed.
For both vertically and horizontally cemented samples, heat
transfers through them vertically, as shown in Fig. 4. Hence, vertical
cementation enhances the particle connectivity along the heat
transfer direction, which enforces the heat transfer in a shorter
path, more straight from top to bottom. Hence, the vertical
cementation results in a higher leff than that in the horizontal di-
rection, as shown in Fig. 10, even though at the same porosity,
which also indicates the importance of microstructure for heat
transfer more than mere porosity. The difference between leff due
to cementation direction also offers a guide to the engineering
application: calcite is desired to be generated along the heat
transfer direction when using MICP to improve the leff of the
ground. As for the three sands either after vertical or horizontal
cementation, the leff of angular sand is still improved to the most
extend, similar to the surface cementation in Section 4.2. Hence,
improving leff in irregular sands by cementation is more
significant.
 W. Fei, G.A. Narsilio / Journal of Rock Mechanics and Geotechnical Engineering 14 (2022) 210e218 217

Fig. 8. leff from the thermal conductance network model versus porosity in cemented sands. Note that the x-axis is inverted to present the increase of solid volume due to
cementation.

Fig. 9. The effect of cementation level on the increase of normalised leff from (a) thermal conductance network model and (b) normalised CN.

5. Conclusions transfer direction can create more preferential heat transfer paths
and thus lead to higher leff than that of the condition when
In order to improve the efficiencies of geothermal systems, cementation is perpendicular to heat transfer.
surface and anisotropic cementations were imposed to three sands
varying in particle shape, aiming to achieve superior effective
thermal conductivity (leff ). A TCNM was implemented to calculate
the leff of the sands. The comparison between leff of uncemented
sands from TCNM, FEM simulation and experiments demonstrates
that leff from TCNM aligns with experimental measurement and
TCNM can overcome the potential overestimation of leff in FEM
simulations. The cementation of the sands in this paper was
simulated by increasing the mineral fraction gradually, thus the
interparticle contact area increases at the same moment. The leff of
the cemented samples from TCNM fall within theoretical results,
showing the robustness of TCNM for simulating granular materials
with large contact areas.
Cementation can increase the interparticle contact area more
efficiently than applying external pre-loading of the ground.
Additionally, a sand with more irregular particles show a larger
enhancement of leff during cementation than more rounded par-
ticle sands. It is also noticeable that cementation along the heat Fig. 10. leff of three sands after directional cementation.
218 W. Fei, G.A. Narsilio / Journal of Rock Mechanics and Geotechnical Engineering 14 (2022) 210e218
Declaration of competing interest
The authors declare that they have no known competing
financial interests or personal relationships that could have
appeared to influence the work reported in this paper.
Acknowledgments
Project ARC DP210100433, and Dr. Anton Maksimenko and
other beam scientists of the Imaging and Medical Beam Line (IMBL)
at the Australian Synchrotron are acknowledged for their support.
References
Asakuma, Y., Asada, M., Kanazawa, Y., Yamamoto, T., 2016. Thermal analysis with
contact resistance of packed bed by a homogenization method. Powder Tech-
nol. 291, 46e51.
Askari, R., Taheri, S., Hejazi, S.H., 2015. Thermal conductivity of granular porous
media: a pore scale modeling approach. AIP Adv. 5 (9), 097106.
ASTM D5334e14, 2014. Standard Test Method for Determination of Thermal Con-
ductivity of Soil and Soft Rock by Thermal Needle Probe Procedure. ASTM In-
ternational, West Conshohocken, PA, USA.
Aurenhammer, F., Klein, R., Lee, D.T., 2013. Voronoi Diagrams and Delaunay Tri-
angulations. World Scientific Publishing Company, Hackensack, NJ, USA.
Bauer, R., Schlunder, E., 1978. Effective radial thermal-conductivity of packings in
gas flow. Part II. Thermal conductivity of packing fraction without gas flow. Int.
Chem. Eng. 18 (2), 189e204.
Chaudhuri, B., Muzzio, F.J., Tomassone, M.S., 2006. Modeling of heat transfer in
granular flow in rotating vessels. Chem. Eng. Sci. 61 (19), 6348e6360.
Comsol Ab, 2020. COMSOL Multiphysics v5.5. http://www.comsol.com.
de Macedo, R.B., Marshall, J.P., Andrade, J.E., 2018. Granular object morphological
generation with genetic algorithms for discrete element simulations. Granul.
Matter 20 (4), 73.
DeVera, A.L., Strieder, W., 1977. Upper and lower bounds on the thermal conduc-
tivity of a random, two-phase material. J. Phys. Chem. 81 (18), 1783e1790.
El Shamy, U., De Leon, O., Wells, R., 2013. Discrete element method study on effect
of shear-induced anisotropy on thermal conductivity of granular soils. Int. J.
GeoMech. 13 (1), 57e64.
Fei, W., Narsilio, G.A., 2020. Network analysis of heat transfer in sands. Comput.
Geotech. 127, 103773.
Fei, W., Narsilio, G.A., Disfani, M.M., 2019b. Impact of three-dimensional sphericity
and roundness on heat transfer in granular materials. Powder Technol. 355,
770e781.
Fei, W., Narsilio, G.A., Disfani, M.M., 2021a. Predicting effective thermal conductivity
in sand using an artificial neural network with multiscale microstructural pa-
rameters. Int. J. Heat Mass Tran. 170, 120997.
Fei, W., Narsilio, G.A., van der Linden, J.H., Disfani, M.M., 2019a. Quantifying the
impact of rigid interparticle structures on heat transfer in granular materials
using networks. Int. J. Heat Mass Tran. 143, 118514.
Fei, W., Narsilio, G.A., van der Linden, J.H., Disfani, M.M., Miao, X., Yang, B., Ashfar, T.,
2021b. X-ray computed tomography images and network data of sands under
compression. Data Brief 36, 107122.
Gostick, J., Aghighi, M., Hinebaugh, J., Tranter, T., Hoeh, M.A., Day, H., Spellacy, B.,
Sharqawy, M.H., Bazylak, A., Burns, A., 2016. Openpnm: a pore network
modeling package. Comput. Sci. Eng. 18 (4), 60e74.
Hashin, Z., Shtrikman, S., 1962. A variational approach to the theory of the effective
magnetic permeability of multiphase materials. J. Appl. Phys. 33 (10), 3125e
3131.
Hill, R., 1965. A self-consistent mechanics of composite materials. J. Mech. Phys.
Solid. 13 (4), 213e222.
Lichtner, P.C., 1988. The quasi-stationary state approximation to coupled mass
transport and fluid-rock interaction in a porous medium. Geochem. Cosmo-
chim. Acta 52 (1), 143e165.
Moscardini, M., Gan, Y., Pupeschi, S., Kamlah, M., 2018. Discrete element method for
effective thermal conductivity of packed pebbles accounting for the smo-
luchowski effect. Fusion Eng. Des. 127, 192e201.
Narsilio, G., Yun, T., Kress, J., Evans, T., 2010a. Hydraulic and thermal conduction
phenomena in soils at the particle-scale: towards realistic fem simulations. IOP
Conf. Ser. Mater. Sci. Eng. 10, 012086.
Narsilio, G.A., Kress, J., Yun, T.S., 2010b. Characterisation of conduction phenomena
in soils at the particle-scale: finite element analyses in conjunction with syn-
thetic 3D imaging. Comput. Geotech. 37 (7e8), 828e836.
Otsu, N., 1979. A threshold selection method from gray-level histograms. IEEE Trans.
Syst. Man Cybern. 9 (1), 62e66.
Persson, B., Albohr, O., Tartaglino, U., Volokitin, A., Tosatti, E., 2005. On the nature of
surface roughness with application to contact mechanics, sealing, rubber fric-
tion and adhesion. J. Phys. Condens. Matter 17 (1), R1eR62.
Sass, J., Lachenbruch, A.H., Munroe, R.J., 1971. Thermal conductivity of rocks from
measurements on fragments and its application to heat-flow determinations.
J. Geophys. Res. 76 (14), 3391e3401.
Shapiro, M., Dudko, V., Royzen, V., Krichevets, Y., Lekhtmakher, S., Grozubinsky, V.,
Shapira, M., Brill, M., 2004. Characterization of powder beds by thermal con-
ductivity: effect of gas pressure on the thermal resistance of particle contact
points. Part. Part. Syst. Char. 21 (4), 268e275.
Tarnawski, V., Leong, W., Gori, F., Buchan, G., Sundberg, J., 2002. Inter-particle
contact heat transfer in soil systems at moderate temperatures. Int. J. Energy
Res. 26 (15), 1345e1358.
Tarnawski, V., Momose, T., Leong, W., 2011. Thermal conductivity of standard sands
II. Saturated conditions. Int. J. Thermophys. 32 (5), 984.
van der Linden, J.H., Narsilio, G.A., Tordesillas, A., 2016. Machine learning frame-
work for analysis of transport through complex networks in porous, granular
media: a focus on permeability. Phys. Rev. E 94 (2), 022904.
van der Linden, J.H., Narsillio, G.A., Tordesillas, A., 2021. Thermal conductance
network model for computerised tomography images of real dry geomaterials.
Comput. Geotech. 136, 104093.
Wang, Z., Zhang, N., Ding, J., Li, Q., Xu, J., 2020. Thermal conductivity of sands
treated with microbially induced calcite precipitation (MICP) and model pre-
diction. Int. J. Heat Mass Tran. 147, 118899.
Wiebicke, M., Andò, E., Herle, I., Viggiani, G., 2017. On the metrology of interparticle
contacts in sand from x-ray tomography images. Meas. Sci. Technol. 28 (12),
124007.
Young, H.D., Freedman, R.A., Sandin, T., Ford, A.L., 1996. University Physics. Addison-
Wesley Reading, Boston, MA, USA.
Yun, T.S., Evans, T.M., 2010. Three-dimensional random network model for thermal
conductivity in particulate materials. Comput. Geotech. 37 (7e8), 991e998.
Yun, T.S., Santamarina, J.C., 2008. Fundamental study of thermal conduction in dry
soils. Granul. Matter 10 (3), 197.
Zhang, N., Zou, H., Zhang, L., Puppala, A.J., Liu, S., Cai, G., 2020. A unified soil thermal
conductivity model based on artificial neural network. Int. J. Therm. Sci. 155,
106414.
Dr. Wenbin Fei is a Research Fellow at the Department of
Infrastructure Engineering, The University of Melbourne,
Australia. He received his PhD degree from the same
University in 2020. Wenbin explores the multi-scale ther-
malehydraulicemechanicalechemical (THMC) processes
in geomaterials with contributions to geothermal engi-
neering, energy structure, and energy waste management
(carbon geological storage and recycled materials in con-
crete and permeable pavement). At particle and pore level,
he developed a platform combining computed tomogra-
phy (CT) technique, advanced image processing algorithm,
discrete element method (DEM), complex network theory
(i.e. graph theory), finite element method (FEM), network
model and machine learning, with the aim to characterise
the microstructure of geomaterials and predict thermal conductivity and permeability
automatically, efficiently and without artificial bias. At the field scale, he used COMO-
SOL Multiphysics and also developed a software AEEA-Coupler which combines
different commercial software (ABAQUS and ECLIPSE) to simulate the THMC problems
in real-size projects. Further information at wenbinfei.github.io.